Upstream the ONNX importer. (#2636)

This is part 1 of 2, which will also include upstreaming the FX
importer. I started with ONNX because it forces some project layout
updates and is more self contained/easier as a first step.

Deviating somewhat from the RFCs on project layout, I made the following
decisions:

* Locating the `onnx_importer.py` into `torch_mlir.extras` as Maks
already has opened up that namespace and it seemed to fit. Better to
have fewer things at that level.
* Setup the build so that the root project only contains MLIR Python and
pure Python deps (like the importers), but this can be augmented with
the `projects/` adding more depending on which features are enabled.
* The default build continues to build everything whereas in
`TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS=1` mode, it builds a
`torch-mlir-core` wheel with the pure contents only.

`onnx_importer.py` and `importer_smoke_test.py` are almost verbatim
copies from SHARK-Turbine. I made some minor local alterations to adapt
to paths and generalize the way they interact with the outer project. I
expect I can copy these back to Turbine verbatim from here. I also
updated the license boilerplate (they have the same license but slightly
different project norms for the headers) but retained the correct
copyright.

Other updates:

* Added the ONNX importer unit test (which also can generate test data)
in lit, conditioned on the availability of the Python `onnx` package. In
a followup once I know everything is stable, I'll add another env var
that the CI can set to always enable this so we know conclusively if
tests pass.
* Moved the ONNX conversion readme to `docs/`.
* Renamed CMake option `TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS` ->
`TORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS` and inverted the sense. Made the
JitIR importer and LTC options `cmake_dependent_options` for robustness.

CMakeLists.txt

@@ -25,6 +25,8 @@ project(torch-mlir LANGUAGES CXX C)
 set(CMAKE_C_STANDARD 11)
 set(CMAKE_CXX_STANDARD 17)
 
+include(CMakeDependentOption)
+
 #-------------------------------------------------------------------------------
 # Project options
 #-------------------------------------------------------------------------------
@@ -43,24 +45,11 @@ endif()
 
 option(TORCH_MLIR_OUT_OF_TREE_BUILD "Specifies an out of tree build" OFF)
 
-# PT1 options.
+# PyTorch native extension gate. If OFF, then no features which depend on
-option(TORCH_MLIR_ENABLE_PROJECT_PT1 "Enables the PyTorch1 project under projects/pt1" OFF)
+# native extensions will be built.
-# TODO: Rename/scope these. They use historic names for now to ease migration
+option(TORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS "Enables PyTorch native extension features" ON)
-# burden.
+cmake_dependent_option(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER "Enables JIT IR Importer" ON TORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS OFF)
-option(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER "Enables JIT IR Importer" ON)
+cmake_dependent_option(TORCH_MLIR_ENABLE_LTC "Enables LTC backend" OFF TORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS OFF)
-option(TORCH_MLIR_ENABLE_LTC "Enables LTC backend" OFF)
-option(TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS "Build Torch dialect MLIR Python bindings but neither JIT IR Importer nor LTC backend" OFF)
-if(TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS)
-  set(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER OFF)
-  set(TORCH_MLIR_ENABLE_LTC OFF)
-endif()
-# Force enable the PT1 project if either the JIT_IR_IMPORTER or LTC is enabled.
-if(NOT TORCH_MLIR_ENABLE_PROJECT_PT1)
-  if(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER OR TORCH_MLIR_ENABLE_LTC)
-    message(STATUS "Enabling projects/pt1 because features requiring it are enabled")
-    set(TORCH_MLIR_ENABLE_PROJECT_PT1 ON)
-  endif()
-endif()
 
 #-------------------------------------------------------------------------------
 # Configure out-of-tree vs in-tree build
@@ -235,4 +224,16 @@ endif()
 # Sub-projects
 #-------------------------------------------------------------------------------
 
+# Sub-projects can bundle additional PyTorch extensions by adding them to this
+# source target. It is typically empty unless if features are enabled.
+if(MLIR_ENABLE_BINDINGS_PYTHON)
+  declare_mlir_python_sources(TorchMLIRPythonTorchExtensionsSources)
+endif()
+
+# Build projects first as it may populate additional Python deps.
 add_subdirectory(projects)
+
+# Finish with top-level Python bindings so it can handle additional deps.
+if(MLIR_ENABLE_BINDINGS_PYTHON)
+  add_subdirectory(python)
+endif()

build_tools/python_deploy/build_linux_packages.sh

@@ -351,7 +351,6 @@ function setup_venv() {
       echo ":::: Using stable dependencies"
       python3 -m pip install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cpu
       python3 -m pip install --no-cache-dir -r /main_checkout/torch-mlir/build-requirements.txt
-      python3 -m pip install --no-cache-dir -r /main_checkout/torch-mlir/test-requirements.txt
       ;;
     *)
       echo "Unrecognized torch version '$torch_version'"
@@ -359,6 +358,7 @@ function setup_venv() {
       ;;
   esac
 
+  python3 -m pip install --no-cache-dir -r /main_checkout/torch-mlir/test-requirements.txt
 }
 
 function build_out_of_tree() {

docs/importers/onnx_importer.md

@@ -3,11 +3,8 @@
 We enable the direct representation of many ONNX features directly in
 the `torch` dialect as `torch.operator` custom ops with names like
 `onnx.{OperatorName}`. The majority of ONNX operators are represented
-with a systematic transformation. See 
+with a systematic transformation. `torch_mlir.extras.onnx_importer`
-[onnx_importer.py](https://github.com/nod-ai/SHARK-Turbine/blob/main/python/shark_turbine/importers/onnx_importer.py)
+for the reference importer which complies with the rules below.
-for the reference importer which complies with the rules below
-(this is planned to be upstreamed to torch-mlir proper in the near
-future).
 
 ## Adding new ONNX operators
 
@@ -26,10 +23,11 @@ are relatively straight-forward to map, following this general procedure:
 * Open the corresponding implementation file `DefaultDomainXtoY.cpp`
   corresponding with the alphabetic sort of the op and add a conversion.
 * Generate successful test cases:
-  * Either run the Turbine importer to produce MLIR output for all
+  * All `onnx_importer.py` tests are dumped to the test temp dir (success
-    ops/models in the ONNX test suite or use a dump that someone has
+    or failure). This is typically located under 
-    generated:
+    `tools/torch-mlir/test/python/onnx_importer/Output`. The `.mlir` files
-      * [2023-Nov-21](https://drive.google.com/file/d/1P6QaRXGnCeApjdjNmykLxWa-yqMmIO-d/view?usp=sharing)
+    under there should provide good variants to drive lit test coverage of
+    conversion.
   * There are often many variants of tests for checking conformance of
     different historic ONNX encodings, but these are often not load bearing
     at the MLIR level.

projects/CMakeLists.txt

@@ -1,7 +1,31 @@
 include(AddMLIRPython)
 
+################################################################################
+# PyTorch
 # Configure PyTorch if we have any features enabled which require it.
+################################################################################
 if(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER OR TORCH_MLIR_ENABLE_LTC)
+
+  if (NOT TORCH_MLIR_USE_INSTALLED_PYTORCH)
+    # Source builds
+    message(STATUS "Building libtorch from source (features depend on it and NOT TORCH_MLIR_USE_INSTALLED_PYTORCH)")
+    set(ENV{TORCH_MLIR_SRC_PYTORCH_REPO} ${TORCH_MLIR_SRC_PYTORCH_REPO})
+    set(ENV{TORCH_MLIR_SRC_PYTORCH_BRANCH} ${TORCH_MLIR_SRC_PYTORCH_BRANCH})
+    set(ENV{TM_PYTORCH_INSTALL_WITHOUT_REBUILD} ${TM_PYTORCH_INSTALL_WITHOUT_REBUILD})
+    set(ENV{MACOSX_DEPLOYMENT_TARGET} ${MACOSX_DEPLOYMENT_TARGET})
+    set(ENV{CMAKE_OSX_ARCHITECTURES} ${CMAKE_OSX_ARCHITECTURES})
+    set(ENV{CMAKE_C_COMPILER_LAUNCHER} ${CMAKE_C_COMPILER_LAUNCHER})
+    set(ENV{CMAKE_CXX_COMPILER_LAUNCHER} ${CMAKE_CXX_COMPILER_LAUNCHER})
+    execute_process(
+            COMMAND ${TORCH_MLIR_SOURCE_DIR}/build_tools/build_libtorch.sh
+            RESULT_VARIABLE _result
+    )
+    if(_result)
+      message(FATAL_ERROR "Failed to run `build_libtorch.sh`")
+    endif()
+    set(TORCH_INSTALL_PREFIX "libtorch")
+  endif()
+
   message(STATUS "Enabling PyTorch C++ dep (features depend on it)")
   include(TorchMLIRPyTorch)
 
@@ -48,6 +72,6 @@ if(TORCH_MLIR_ENABLE_LTC)
 endif()
 
 # Include overall PT1 project.
-if(TORCH_MLIR_ENABLE_PROJECT_PT1)
+if(TORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS)
   add_subdirectory(pt1)
 endif()

projects/pt1/python/CMakeLists.txt

@@ -7,79 +7,22 @@ set(CMAKE_PLATFORM_NO_VERSIONED_SONAME ON)
 # argument.
 set(TORCH_MLIR_PYTHON_ROOT_DIR "${CMAKE_CURRENT_SOURCE_DIR}/torch_mlir")
 
-
 # We vendor our own MLIR instance in the `torch_mlir` namespace.
 add_compile_definitions("MLIR_PYTHON_PACKAGE_PREFIX=torch_mlir.")
 
-################################################################################
+# ################################################################################
-# PyTorch
+# # Sources
-################################################################################
+# ################################################################################
 
-if (NOT TORCH_MLIR_USE_INSTALLED_PYTORCH)
+declare_mlir_python_sources(TorchMLIRPythonSources.TopLevel
-  # Source builds
-  set(ENV{TORCH_MLIR_SRC_PYTORCH_REPO} ${TORCH_MLIR_SRC_PYTORCH_REPO})
-  set(ENV{TORCH_MLIR_SRC_PYTORCH_BRANCH} ${TORCH_MLIR_SRC_PYTORCH_BRANCH})
-  set(ENV{TM_PYTORCH_INSTALL_WITHOUT_REBUILD} ${TM_PYTORCH_INSTALL_WITHOUT_REBUILD})
-  set(ENV{MACOSX_DEPLOYMENT_TARGET} ${MACOSX_DEPLOYMENT_TARGET})
-  set(ENV{CMAKE_OSX_ARCHITECTURES} ${CMAKE_OSX_ARCHITECTURES})
-  set(ENV{CMAKE_C_COMPILER_LAUNCHER} ${CMAKE_C_COMPILER_LAUNCHER})
-  set(ENV{CMAKE_CXX_COMPILER_LAUNCHER} ${CMAKE_CXX_COMPILER_LAUNCHER})
-  execute_process(
-          COMMAND ${TORCH_MLIR_SOURCE_DIR}/build_tools/build_libtorch.sh
-          RESULT_VARIABLE _result
-  )
-  if(_result)
-    message(FATAL_ERROR "Failed to run `build_libtorch.sh`")
-  endif()
-  set(TORCH_INSTALL_PREFIX "libtorch")
-endif()
-
-################################################################################
-# Sources
-################################################################################
-
-declare_mlir_python_sources(TorchMLIRPythonSources)
-declare_mlir_python_sources(TorchMLIRPythonExtensions)
-
-if (NOT TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS)
-  declare_mlir_python_sources(TorchMLIRPythonSources.TopLevel
-    ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
-    ADD_TO_PARENT TorchMLIRPythonSources
-    SOURCES
-      __init__.py
-      _dynamo_fx_importer.py
-      compiler_utils.py
-      dynamo.py
-      _version.py
-  )
-endif()
-
-declare_mlir_python_sources(TorchMLIRPythonSources.Dialects
   ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
-  ADD_TO_PARENT TorchMLIRPythonSources
+  ADD_TO_PARENT TorchMLIRPythonTorchExtensionsSources
-)
-
-declare_mlir_dialect_python_bindings(
-  ADD_TO_PARENT TorchMLIRPythonSources.Dialects
-  ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
-  TD_FILE dialects/TorchBinding.td
-  SOURCES dialects/torch/__init__.py
-  DIALECT_NAME torch
-)
-
-################################################################################
-# Extensions
-################################################################################
-
-declare_mlir_python_extension(TorchMLIRPythonExtensions.Main
-  MODULE_NAME _torchMlir
-  ADD_TO_PARENT TorchMLIRPythonExtensions
   SOURCES
-    TorchMLIRModule.cpp
+    __init__.py
-  EMBED_CAPI_LINK_LIBS
+    _dynamo_fx_importer.py
-    TorchMLIRCAPI
+    compiler_utils.py
-  PRIVATE_LINK_LIBS
+    dynamo.py
-    LLVMSupport
+    _version.py
 )
 
 ################################################################################
@@ -110,56 +53,23 @@ endif()
 
 # add_subdirectory(torch_mlir/_torch_mlir_custom_op_example)
 
-################################################################################
-# Generate packages and shared library
-# Downstreams typically will not use these, but they are useful for local
-# testing.
-################################################################################
-
-set(_source_components
-  # TODO: Core is now implicitly building/registering all dialects, increasing
-  # build burden by ~5x. Make it stop.
-  # TODO: Reduce dependencies. We need ExecutionEngine and a bunch of passes
-  # for the reference backend, but logically they can be separate. But seemingly
-  # the only way to handle that is to create a separate mlir python package
-  # tree, which seems excessive.
-  MLIRPythonSources
-  MLIRPythonExtension.Core
-  MLIRPythonExtension.RegisterEverything
-  TorchMLIRPythonSources
-  TorchMLIRPythonExtensions
-)
-
-add_mlir_python_common_capi_library(TorchMLIRAggregateCAPI
-  INSTALL_COMPONENT TorchMLIRPythonModules
-  INSTALL_DESTINATION python_packages/torch_mlir/torch_mlir/_mlir_libs
-  OUTPUT_DIRECTORY "${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/_mlir_libs"
-  RELATIVE_INSTALL_ROOT "../../../.."
-  DECLARED_SOURCES ${_source_components}
-)
-
-add_mlir_python_modules(TorchMLIRPythonModules
-  ROOT_PREFIX "${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir"
-  INSTALL_PREFIX "python_packages/torch_mlir/torch_mlir"
-  DECLARED_SOURCES ${_source_components}
-  COMMON_CAPI_LINK_LIBS
-    TorchMLIRAggregateCAPI
-  )
-
 # TODO: Find a cleaner way to do this.
 # Can we build the JIT IR importer with `declare_mlir_python_extension`?
 # Then it would "just work".
 if(TORCH_MLIR_ENABLE_JIT_IR_IMPORTER)
-  add_dependencies(TorchMLIRPythonModules TorchMLIRJITIRImporter)
+  add_dependencies(TorchMLIRPythonTorchExtensionsSources 
-  add_dependencies(TorchMLIRPythonModules TorchMLIRJITIRImporterPybind)
+    TorchMLIRJITIRImporter
-  # Build the E2E Tests (which depend on the JIT IR importer now).
+    TorchMLIRJITIRImporterPybind
-  add_dependencies(TorchMLIRPythonModules TorchMLIRE2ETestPythonModules)
+    TorchMLIRE2ETestPythonModules
+  )
 endif()
 
 if(TORCH_MLIR_ENABLE_LTC)
   # Add Torch-MLIR LTC backend as dependency
-  add_dependencies(TorchMLIRPythonModules torch_mlir_ltc_backend)
+  add_dependencies(TorchMLIRPythonTorchExtensionsSources 
-  add_dependencies(TorchMLIRPythonModules reference_lazy_backend)
+    torch_mlir_ltc_backend
+    reference_lazy_backend
+  )
 endif()
 
 add_subdirectory(test)

projects/pt1/python/torch_mlir/jit_ir_importer/CMakeLists.txt

@@ -4,9 +4,9 @@
 
 ## Declare the sources of the Python module.
 
-declare_mlir_python_sources(TorchMLIRPythonSources.JitIRImporter
+declare_mlir_python_sources(TorchMLIRPythonTorchExtensionsSources.JitIRImporter
   ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
-  ADD_TO_PARENT TorchMLIRPythonSources
+  ADD_TO_PARENT TorchMLIRPythonTorchExtensionsSources
   SOURCES_GLOB
     jit_ir_importer/*.py
 )

python/CMakeLists.txt

@@ -0,0 +1,94 @@
+# Disables generation of "version soname" (i.e. libFoo.so.<version>), which
+# causes pure duplication as part of Python wheels.
+set(CMAKE_PLATFORM_NO_VERSIONED_SONAME ON)
+
+# The directory at which the Python import tree begins.
+# See documentation for `declare_mlir_python_sources`'s ROOT_DIR
+# argument.
+set(TORCH_MLIR_PYTHON_ROOT_DIR "${CMAKE_CURRENT_SOURCE_DIR}/torch_mlir")
+
+
+# We vendor our own MLIR instance in the `torch_mlir` namespace.
+add_compile_definitions("MLIR_PYTHON_PACKAGE_PREFIX=torch_mlir.")
+
+################################################################################
+# Sources
+################################################################################
+
+declare_mlir_python_sources(TorchMLIRPythonSources)
+declare_mlir_python_sources(TorchMLIRPythonExtensions)
+
+declare_mlir_python_sources(TorchMLIRPythonSources.Dialects
+  ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
+  ADD_TO_PARENT TorchMLIRPythonSources
+)
+
+declare_mlir_dialect_python_bindings(
+  ADD_TO_PARENT TorchMLIRPythonSources.Dialects
+  ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
+  TD_FILE dialects/TorchBinding.td
+  SOURCES dialects/torch/__init__.py
+  DIALECT_NAME torch
+)
+
+declare_mlir_python_sources(TorchMLIRPythonSources.Importers
+  ROOT_DIR "${TORCH_MLIR_PYTHON_ROOT_DIR}"
+  ADD_TO_PARENT TorchMLIRPythonSources
+  SOURCES
+    extras/onnx_importer.py
+)
+
+################################################################################
+# Extensions
+################################################################################
+
+declare_mlir_python_extension(TorchMLIRPythonExtensions.Main
+  MODULE_NAME _torchMlir
+  ADD_TO_PARENT TorchMLIRPythonExtensions
+  SOURCES
+    TorchMLIRModule.cpp
+  EMBED_CAPI_LINK_LIBS
+    TorchMLIRCAPI
+  PRIVATE_LINK_LIBS
+    LLVMSupport
+)
+
+################################################################################
+# Generate packages and shared library
+# Downstreams typically will not use these, but they are useful for local
+# testing.
+################################################################################
+
+set(_source_components
+  # TODO: Core is now implicitly building/registering all dialects, increasing
+  # build burden by ~5x. Make it stop.
+  # TODO: Reduce dependencies. We need ExecutionEngine and a bunch of passes
+  # for the reference backend, but logically they can be separate. But seemingly
+  # the only way to handle that is to create a separate mlir python package
+  # tree, which seems excessive.
+  MLIRPythonSources
+  MLIRPythonExtension.Core
+  MLIRPythonExtension.RegisterEverything
+  TorchMLIRPythonSources
+  TorchMLIRPythonExtensions
+
+  # Sources related to optional Torch extension dependent features. Typically
+  # empty unless if project features are enabled.
+  TorchMLIRPythonTorchExtensionsSources
+)
+
+add_mlir_python_common_capi_library(TorchMLIRAggregateCAPI
+  INSTALL_COMPONENT TorchMLIRPythonModules
+  INSTALL_DESTINATION python_packages/torch_mlir/torch_mlir/_mlir_libs
+  OUTPUT_DIRECTORY "${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/_mlir_libs"
+  RELATIVE_INSTALL_ROOT ".."
+  DECLARED_SOURCES ${_source_components}
+)
+
+add_mlir_python_modules(TorchMLIRPythonModules
+  ROOT_PREFIX "${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir"
+  INSTALL_PREFIX "python_packages/torch_mlir/torch_mlir"
+  DECLARED_SOURCES ${_source_components}
+  COMMON_CAPI_LINK_LIBS
+    TorchMLIRAggregateCAPI
+  )

python/torch_mlir/extras/onnx_importer.py

@@ -0,0 +1,607 @@
+# Based on code Copyright (c) Advanced Micro Devices, Inc.
+#
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Also available under a BSD-style license. See LICENSE.
+
+"""Imports ONNX graphs to `torch` dialect ops.
+
+See documentation:
+    https://github.com/llvm/torch-mlir/blob/main/docs/importers/onnx_importer.md
+
+This file is distributed/forked verbatim into various downstream projects, and
+it must abide by several rules above and beyond the rest of the codebase:
+    - It must be standalone, only depending on:
+        - `onnx`
+        - `..ir` relative imports to the main IR directory
+        - `..dialects.func` relative import to the `func` dialect (TODO:
+           we are looking to eliminate this dep).
+        - Python standard library
+    - It does not directly use the ODS generated `torch` dialect Python
+      wrappers. This allows it to be used in contexts that only build a C++
+      compiler with minimal IR Python bindings.
+    - It is intended as an enabler for full onnx compilation, only handling
+      the import from ONNX -> the `torch` dialect. Testing, full pipelines,
+      and utilities belong elsewhere.
+"""
+
+try:
+    import onnx
+except ModuleNotFoundError as e:
+    raise ModuleNotFoundError(
+        "The onnx package (`pip install onnx`) is required to use the onnx importer"
+    ) from e
+
+from typing import Optional
+
+from dataclasses import dataclass
+
+import numpy as np
+
+from ..ir import (
+    ArrayAttr,
+    Attribute,
+    Block,
+    Context,
+    DenseElementsAttr,
+    DenseResourceElementsAttr,
+    DictAttr,
+    FloatAttr,
+    BF16Type,
+    ComplexType,
+    F16Type,
+    F32Type,
+    F64Type,
+    Float8E4M3FNType,
+    Float8E5M2FNUZType,
+    Float8E5M2Type,
+    FunctionType,
+    InsertionPoint,
+    IntegerAttr,
+    IntegerType,
+    MLIRError,
+    RankedTensorType,
+    Location,
+    Module,
+    Operation,
+    StringAttr,
+    Type as IrType,
+    Value,
+)
+
+from ..dialects import (
+    func as func_dialect,
+)
+
+@dataclass
+class Config:
+    """Various configuration settings for the importer."""
+
+    # Ancient ONNX exporters would often add a model input for anything that
+    # might be mutable, providing an initializer for it as well. More modern
+    # tools tools realized this is a really bad idea for a lot of reasons.
+    # We choose to assume more recent norms, even if encountering older
+    # models. Setting this to False probably won't do what you want but
+    # should produce interesting errors to waste your time deciphering.
+    # We mainly use it as a way to document in the code that we are
+    # making an assumption.
+    elide_initialized_inputs: bool = True
+
+
+class ModelInfo:
+    """Top-level accounting and accessors for an ONNX model."""
+
+    def __init__(self, model_proto: onnx.ModelProto, *, config: Config = Config()):
+        self.config = config
+        self.model_proto = model_proto
+        assert model_proto.graph, "Model must contain a main Graph"
+        self.main_graph = GraphInfo(self, model_proto.graph)
+
+    def create_module(self, context: Optional[Context] = None) -> Operation:
+        if not context:
+            context = Context()
+        module_op = Module.create(Location.unknown(context)).operation
+        # TODO: Populate module level metadata from the ModelProto
+        return module_op
+
+
+class GraphInfo:
+    """Information about a Graph within a model."""
+
+    def __init__(self, model_info: ModelInfo, graph_proto: onnx.GraphProto):
+        self.model_info = model_info
+        self.graph_proto = graph_proto
+        self.initializer_map: dict[str, onnx.TensorProto] = {
+            n.name: n for n in graph_proto.initializer
+        }
+        self.value_info_map: dict[str, onnx.ValueInfoProto] = {
+            n.name: n for n in graph_proto.value_info
+        }
+        self.declared_input_map: dict[str, onnx.ValueInfoProto] = {
+            n.name: n for n in graph_proto.input
+        }
+        self.output_map: dict[str, onnx.ValueInfoProto] = {
+            n.name: n for n in graph_proto.output
+        }
+
+        # Generate the effective input map, which for old models can be a
+        # subset of the input map.
+        if model_info.config.elide_initialized_inputs:
+            self.input_map = {
+                k: v
+                for k, v in self.declared_input_map.items()
+                if k not in self.initializer_map
+            }
+        else:
+            self.input_map = self.declared_input_map
+            illegal_input_keys = self.input_map.keys() - (
+                self.input_map.keys() - self.initializer_map.keys()
+            )
+            assert self.input_map.keys().isdisjoint(self.initializer_map.keys()), (
+                f"When not in elide_initialized_inputs=True, we expect inputs to not "
+                f"have an initial value (got {illegal_input_keys})."
+            )
+
+    def find_type_proto_for_name(self, name: str) -> onnx.TypeProto:
+        # Node outputs don't typically have type information, but shape inference
+        # will associate them in the value_info. If not there, it may be a
+        # graph output, which must have type information.
+        value_info = self.value_info_map.get(name) or self.output_map.get(name)
+        if value_info is not None:
+            return value_info.type
+        raise OnnxImportError(
+            f"No type information associated with '{name}'. Run shape inference?"
+        )
+
+
+class OnnxImportError(Exception):
+    ...
+
+
+class NodeImporter:
+    """Imports graph nodes into MLIR.
+
+    Typically, the top level graph will be imported into a func whereas dependent
+    graphs may just be imported with references to pre-existing values.
+
+    Note that ONNX requires that graphs be sorted topologically and free of cycles,
+    so we don't take any special steps to order them for dominance.
+    """
+
+    __slots__ = [
+        "_c",
+        "_cc",
+        "_gi",
+        "_p",
+        "_b",
+        "_nv_map",
+    ]
+
+    def __init__(
+        self,
+        graph_info: GraphInfo,
+        *,
+        parent_op: Operation,
+        block: Block,
+        context_cache: "ContextCache",
+    ):
+        self._c = parent_op.context
+        self._cc = context_cache
+        self._gi = graph_info
+        self._p = parent_op
+        self._b = block
+        self._nv_map: dict[str, Value] = {}
+
+    @classmethod
+    def define_function(
+        cls, graph_info: GraphInfo, module_op: Operation
+    ) -> "NodeImporter":
+        cc = ContextCache(module_op.context)
+        with module_op.context, Location.name(f"graph:{graph_info.graph_proto.name}"):
+            body = module_op.regions[0].blocks[0]
+            func_name = graph_info.graph_proto.name
+            input_types = [
+                cc.type_proto_to_type(inp.type) for inp in graph_info.input_map.values()
+            ]
+            output_types = [
+                cc.type_proto_to_type(out.type)
+                for out in graph_info.output_map.values()
+            ]
+            ftype = FunctionType.get(input_types, output_types)
+            func_op = func_dialect.FuncOp(func_name, ftype, ip=InsertionPoint(body))
+            block = func_op.add_entry_block(
+                [Location.name(k) for k in graph_info.input_map.keys()]
+            )
+        imp = NodeImporter(graph_info, parent_op=func_op, block=block, context_cache=cc)
+        for node_name, input_value in zip(graph_info.input_map.keys(), block.arguments):
+            imp._nv_map[node_name] = input_value
+        imp._populate_graph_attrs(func_op)
+        return imp
+
+    def _populate_graph_attrs(self, container_op: Operation):
+        """Populates graph level meta attributes on the given container op."""
+        m = self._gi.model_info.model_proto
+        with container_op.context:
+            i64_type = IntegerType.get_signed(64)
+            default_opset_version = 0
+            opset_versions: dict[str, IntegerAttr] = {}
+            for opset_import in m.opset_import:
+                if opset_import.domain:
+                    opset_versions[opset_import.domain] = IntegerAttr.get(
+                        i64_type, opset_import.version
+                    )
+                else:
+                    default_opset_version = opset_import.version
+            if default_opset_version:
+                container_op.attributes[
+                    "torch.onnx_meta.opset_version"
+                ] = IntegerAttr.get(i64_type, default_opset_version)
+            if opset_versions:
+                container_op.attributes[
+                    "torch.onnx_meta.opset_versions"
+                ] = DictAttr.get(opset_versions)
+            container_op.attributes["torch.onnx_meta.ir_version"] = IntegerAttr.get(
+                IntegerType.get_signed(64), m.ir_version
+            )
+            container_op.attributes["torch.onnx_meta.producer_name"] = StringAttr.get(
+                m.producer_name
+            )
+            container_op.attributes[
+                "torch.onnx_meta.producer_version"
+            ] = StringAttr.get(m.producer_version)
+
+    def import_all(self):
+        """Imports all nodes topologically."""
+        # TODO: Consider pulling in initializers on demand since there can be so
+        # much unused crap.
+        for init in self._gi.initializer_map.values():
+            self.import_initializer(init)
+        for node in self._gi.graph_proto.node:
+            self.import_node(node)
+
+        outputs = []
+        for output_name in self._gi.output_map.keys():
+            try:
+                outputs.append(self._nv_map[output_name])
+            except KeyError:
+                raise OnnxImportError(
+                    f"Non topologically produced ONNX graph output '{output_name}'"
+                )
+        with InsertionPoint(self._b), Location.unknown():
+            func_dialect.ReturnOp(outputs)
+
+    def import_node(self, node: onnx.NodeProto):
+        with InsertionPoint(self._b), Location.name(node.name):
+            op_type = node.op_type
+            # Handle special op types that materialize to non-op IR constructs.
+            special_key = f"_handle_node_{op_type}"
+            if hasattr(self, special_key):
+                getattr(self, special_key)(node)
+                return
+
+            # General node import.
+            input_values = []
+            for input_name in node.input:
+                try:
+                    input_values.append(self._nv_map[input_name])
+                except KeyError:
+                    raise OnnxImportError(
+                        f"Non topologically produced ONNX node input '{input_name}': {node}"
+                    )
+
+            output_names = list(node.output)
+            output_types = [
+                self._cc.type_proto_to_type(self._gi.find_type_proto_for_name(n))
+                for n in output_names
+            ]
+
+            # TODO: Attributes.
+            attrs = {
+                "name": StringAttr.get(f"onnx.{op_type}"),
+            }
+            self.import_attributes(node.attribute, attrs)
+            custom_op = Operation.create(
+                name="torch.operator",
+                results=output_types,
+                operands=input_values,
+                attributes=attrs,
+            )
+            for output_name, output_value in zip(output_names, custom_op.results):
+                self._nv_map[output_name] = output_value
+
+    def import_attributes(
+        self, onnx_attrs: list[onnx.AttributeProto], attrs: dict[str, Attribute]
+    ):
+        for onnx_attr in onnx_attrs:
+            attr_type = onnx_attr.type
+            if attr_type not in ATTRIBUTE_TYPE_HANDLERS:
+                raise OnnxImportError(
+                    f"Unhandled ONNX attribute type code {attr_type}: {onnx_attr}"
+                )
+            handler = ATTRIBUTE_TYPE_HANDLERS[attr_type]
+            if handler is None:
+                # Active skip.
+                continue
+            elif handler is False:
+                # Active error.
+                raise OnnxImportError(
+                    f"ONNX importer does not support generic node attribute type {attr_type}. "
+                    f"This likely means that this is a special node which requires specific "
+                    f"handling in the importer: {onnx_attr}"
+                )
+            attrs[f"torch.onnx.{onnx_attr.name}"] = handler(onnx_attr, self._cc)
+
+    def import_initializer(self, initializer: onnx.TensorProto) -> Value:
+        with InsertionPoint(self._b), Location.name(initializer.name):
+            value_attr = self._cc.tensor_proto_to_attr(initializer)
+            vtensor_type = self._cc.tensor_proto_to_type(initializer)
+            literal_op = Operation.create(
+                name="torch.vtensor.literal",
+                results=[vtensor_type],
+                attributes={"value": value_attr},
+            )
+            self._nv_map[initializer.name] = literal_op.result
+        return literal_op.result
+
+    def _get_immediate_tensor(self, name: str) -> np.array:
+        try:
+            initializer = self._gi.initializer_map[name]
+        except KeyError:
+            raise OnnxImportError(
+                f"An immediate value for '{name}' was required but it is dynamically produced."
+            )
+        try:
+            dtype = ELEM_TYPE_TO_NUMPY_DTYPE[initializer.data_type]
+        except KeyError:
+            raise OnnxImportError(
+                f"Unknown ONNX tensor element type to numpy dtype mapping: {initializer.data_type}"
+            )
+        raw_data = initializer.raw_data
+        if raw_data:
+            return np.frombuffer(raw_data, dtype=dtype).reshape(tuple(initializer.dims))
+        else:
+            raise OnnxImportError(
+                f"Unhandled ONNX TensorProto immediate data: {initializer}"
+            )
+
+    def _handle_node_ConstantOfShape(self, node: onnx.NodeProto):
+        # This op is special: It has an input of the shape, and in full generality
+        # could involve eager production of constants of variable size. In
+        # practice, the DNN profile for ONNX makes this very difficult to do
+        # and we hard-assert that the input can be resolved to an immediate
+        # value.
+        assert len(node.input) == 1
+        assert len(node.output) == 1
+        shape = self._get_immediate_tensor(node.input[0]).astype(np.int64)
+        value_proto = _get_attr(node, "value")
+        assert value_proto.type == onnx.AttributeProto.AttributeType.TENSOR
+        tensor_proto = value_proto.t
+        element_type = self._cc.tensor_element_type(tensor_proto.data_type)
+        vtensor_type = self._cc.get_vtensor_type(tuple(shape), element_type)
+        assert len(tensor_proto.dims) == 1 and tensor_proto.dims[0] == 1
+        try:
+            cb = ELEM_TYPE_SPLAT_TENSOR_PROTO_CB[tensor_proto.data_type]
+        except KeyError:
+            raise OnnxImportError(
+                f"Unhandled splat type for ConstantOfShape: {node} (possible missing mapping in ELEM_TYPE_SPLAT_TENSOR_PROTO_CB)"
+            )
+        value_attr = cb(tensor_proto, tuple(shape))
+        literal_op = Operation.create(
+            name="torch.vtensor.literal",
+            results=[vtensor_type],
+            attributes={"value": value_attr},
+        )
+        self._nv_map[node.output[0]] = literal_op.result
+
+
+class ContextCache:
+    """Caches per-context lookups of various things."""
+
+    __slots__ = [
+        "_c",
+        "_elem_type_map",
+        "_vtensor_type_map",
+    ]
+
+    def __init__(self, context: Context):
+        self._c = context
+        self._elem_type_map: dict[int, IrType] = {}
+        self._vtensor_type_map: dict[tuple[tuple[Optional[int]], IrType], IrType] = {}
+
+    def tensor_element_type(self, elem_type: int) -> IrType:
+        t = self._elem_type_map.get(elem_type)
+        if t is None:
+            try:
+                with self._c:
+                    t = ELEM_TYPE_TO_IR_TYPE_CB[elem_type]()
+            except KeyError:
+                raise OnnxImportError(f"Unknown ONNX tensor element type: {elem_type}")
+            self._elem_type_map[elem_type] = t
+        return t
+
+    def get_vtensor_type(
+        self, dims: tuple[Optional[int]], element_type: IrType
+    ) -> IrType:
+        key = (dims, element_type)
+        t = self._vtensor_type_map.get(key)
+        if t is None:
+            shape_asm = ",".join("?" if d is None else str(d) for d in dims)
+            asm = f"!torch.vtensor<[{shape_asm}],{str(element_type)}>"
+            try:
+                t = IrType.parse(asm, context=self._c)
+            except MLIRError as e:
+                raise OnnxImportError(
+                    f"Unparseable torch type (MLIR asm format bug?): {asm}"
+                ) from e
+            self._vtensor_type_map[key] = t
+        return t
+
+    def tensor_proto_to_type(self, tp: onnx.TensorProto) -> IrType:
+        element_type = self.tensor_element_type(tp.data_type)
+        return self.get_vtensor_type(tuple(tp.dims), element_type)
+
+    def tensor_proto_to_builtin_type(self, tp: onnx.TensorProto) -> IrType:
+        element_type = self.tensor_element_type(tp.data_type)
+        # TODO: Fixme upstream: RankedTensorType.get should not require a location.
+        with Location.unknown():
+            return RankedTensorType.get(tuple(tp.dims), element_type)
+
+    def type_proto_to_type(self, tp: onnx.TypeProto) -> IrType:
+        if tp.tensor_type:
+            tt = tp.tensor_type
+            if not tt.shape:
+                raise OnnxImportError(
+                    f"Unsupported Tensor type without shape (run shape inference?): {tp}"
+                )
+            element_type = self.tensor_element_type(tt.elem_type)
+            dims = tuple(
+                (d.dim_value if not d.dim_param else None) for d in tt.shape.dim
+            )
+            return self.get_vtensor_type(dims, element_type)
+        else:
+            # TODO: Others if ever needed. Or we consider ourselves DNN-only.
+            # See TypeProto: sequence_type, map_type, optional_type, sparse_tensor_type.
+            raise OnnxImportError(f"Unsupported ONNX TypeProto: {tp}")
+
+    def tensor_proto_to_attr(self, tp: onnx.TensorProto) -> Attribute:
+        tensor_type = self.tensor_proto_to_builtin_type(tp)
+        if tp.HasField("raw_data"):
+            # Conveniently, DenseResourceElementsAttr shares the raw data
+            # format. We just give it maximum numeric alignment.
+            return DenseResourceElementsAttr.get_from_buffer(
+                tp.raw_data, tp.name, tensor_type, alignment=8
+            )
+        else:
+            # We have to do a data type specific instantiation from proto fields.
+            # Since this is typically used for small tensor constants, we instantiate
+            # as a DenseElementsAttr.
+            handler = ELEM_TYPE_INLINE_TENSOR_PROTO_CB.get(tp.data_type)
+            if handler is None:
+                raise OnnxImportError(f"Unhandled ONNX TensorProto data: {tp}")
+            return handler(tp)
+
+
+ELEM_TYPE_TO_IR_TYPE_CB = {
+    onnx.TensorProto.DataType.FLOAT: lambda: F32Type.get(),
+    onnx.TensorProto.DataType.UINT8: lambda: IntegerType.get_unsigned(8),
+    onnx.TensorProto.DataType.INT8: lambda: IntegerType.get_signed(8),
+    onnx.TensorProto.DataType.UINT16: lambda: IntegerType.get_unsigned(16),
+    onnx.TensorProto.DataType.INT16: lambda: IntegerType.get_signed(16),
+    onnx.TensorProto.DataType.INT32: lambda: IntegerType.get_signed(32),
+    onnx.TensorProto.DataType.INT64: lambda: IntegerType.get_signed(64),
+    onnx.TensorProto.DataType.BOOL: lambda: IntegerType.get_signless(1),
+    onnx.TensorProto.DataType.FLOAT16: lambda: F16Type.get(),
+    onnx.TensorProto.DataType.DOUBLE: lambda: F64Type.get(),
+    onnx.TensorProto.DataType.UINT32: lambda: IntegerType.get_unsigned(32),
+    onnx.TensorProto.DataType.UINT64: lambda: IntegerType.get_unsigned(64),
+    onnx.TensorProto.DataType.COMPLEX64: lambda: ComplexType.get(F32Type.get()),
+    onnx.TensorProto.DataType.COMPLEX128: lambda: ComplexType.get(F64Type.get()),
+    onnx.TensorProto.DataType.BFLOAT16: lambda: BF16Type.get(),
+    onnx.TensorProto.DataType.FLOAT8E4M3FN: lambda: Float8E4M3FNType.get(),
+    onnx.TensorProto.DataType.FLOAT8E4M3FNUZ: lambda: Float8E5M2FNUZType.get(),
+    onnx.TensorProto.DataType.FLOAT8E5M2: lambda: Float8E5M2Type.get(),
+    onnx.TensorProto.DataType.FLOAT8E5M2FNUZ: lambda: Float8E5M2FNUZType.get(),
+    # Ommitted: STRING,
+}
+
+ELEM_TYPE_SPLAT_TENSOR_PROTO_CB = {
+    onnx.TensorProto.DataType.FLOAT: lambda tp, shape: DenseElementsAttr.get_splat(
+        RankedTensorType.get(shape, F32Type.get()), FloatAttr.get_f32(tp.float_data[0])
+    ),
+    # TODO: All the rest from ELEM_TYPE_TO_IR_TYPE_CB
+}
+
+# Mapping of TensorProto.DataType to lambda TensorProto, returning a DenseElementsAttr
+# of the builtin tensor type for cases where the tensor data is inlined as typed
+# values instead of raw_data.
+ELEM_TYPE_INLINE_TENSOR_PROTO_CB = {
+    onnx.TensorProto.DataType.FLOAT: lambda tp: DenseElementsAttr.get(
+        np.asarray(tp.float_data, dtype=np.float32).reshape(tp.dims), signless=False
+    ),
+    onnx.TensorProto.DataType.INT32: lambda tp: DenseElementsAttr.get(
+        np.asarray(tp.int32_data, dtype=np.int32).reshape(tp.dims), signless=False
+    ),
+    onnx.TensorProto.DataType.INT64: lambda tp: DenseElementsAttr.get(
+        np.asarray(tp.int64_data, dtype=np.int64).reshape(tp.dims), signless=False
+    ),
+    onnx.TensorProto.DataType.DOUBLE: lambda tp: DenseElementsAttr.get(
+        np.asarray(tp.double_data, dtype=np.float64).reshape(tp.dims)
+    ),
+    onnx.TensorProto.DataType.UINT32: lambda tp: DenseElementsAttr.get(
+        # Special case. See proto
+        np.asarray(tp.uint64_data, dtype=np.uint32).reshape(tp.dims),
+        signless=False,
+    ),
+    onnx.TensorProto.DataType.UINT64: lambda tp: DenseElementsAttr.get(
+        np.asarray(tp.uint64_data, dtype=np.uint64).reshape(tp.dims), signless=False
+    )
+    # Intentionally unsupported: STRING
+}
+
+ELEM_TYPE_TO_NUMPY_DTYPE = {
+    onnx.TensorProto.DataType.FLOAT: np.float32,
+    onnx.TensorProto.DataType.UINT8: np.uint8,
+    onnx.TensorProto.DataType.INT8: np.int8,
+    onnx.TensorProto.DataType.UINT16: np.uint16,
+    onnx.TensorProto.DataType.INT16: np.int16,
+    onnx.TensorProto.DataType.INT32: np.int32,
+    onnx.TensorProto.DataType.INT64: np.int64,
+    onnx.TensorProto.DataType.BOOL: np.bool_,
+    onnx.TensorProto.DataType.FLOAT16: np.float16,
+    onnx.TensorProto.DataType.DOUBLE: np.float64,
+    onnx.TensorProto.DataType.UINT32: np.uint32,
+    onnx.TensorProto.DataType.UINT64: np.uint64,
+    onnx.TensorProto.DataType.COMPLEX64: np.complex64,
+    onnx.TensorProto.DataType.COMPLEX128: np.complex128,
+    # onnx.TensorProto.DataType.BFLOAT16:
+    # onnx.TensorProto.DataType.FLOAT8E4M3FN:
+    # onnx.TensorProto.DataType.FLOAT8E4M3FNUZ:
+    # onnx.TensorProto.DataType.FLOAT8E5M2:
+    # onnx.TensorProto.DataType.FLOAT8E5M2FNUZ:
+    # Ommitted: STRING,
+}
+
+# Mapping of AttributeType code to one of:
+#   None: Ignore attribute and do not output to MLIR
+#   False: Error if an attribute of this type is present
+#   lambda a:AttributeProto, cc: ContextCache that returns an MLIR Attribute
+ATTRIBUTE_TYPE_HANDLERS = {
+    onnx.AttributeProto.AttributeType.UNDEFINED: False,
+    onnx.AttributeProto.AttributeType.FLOAT: lambda a, cc: FloatAttr.get(
+        F32Type.get(), a.f
+    ),
+    onnx.AttributeProto.AttributeType.INT: lambda a, cc: IntegerAttr.get(
+        IntegerType.get_signed(64), a.i
+    ),
+    onnx.AttributeProto.AttributeType.STRING: lambda a, cc: StringAttr.get(a.s),
+    onnx.AttributeProto.AttributeType.TENSOR: lambda a, cc: cc.tensor_proto_to_attr(
+        a.t
+    ),
+    onnx.AttributeProto.AttributeType.GRAPH: False,
+    onnx.AttributeProto.AttributeType.SPARSE_TENSOR: False,
+    onnx.AttributeProto.AttributeType.TYPE_PROTO: False,
+    onnx.AttributeProto.AttributeType.FLOATS: lambda a, cc: ArrayAttr.get(
+        [FloatAttr.get(F32Type.get(), f) for f in a.floats]
+    ),
+    onnx.AttributeProto.AttributeType.INTS: lambda a, cc: ArrayAttr.get(
+        [IntegerAttr.get(IntegerType.get_signed(64), i) for i in a.ints]
+    ),
+    onnx.AttributeProto.AttributeType.STRINGS: lambda a, cc: ArrayAttr.get(
+        [StringAttr.get(s) for s in a.strings]
+    ),
+    onnx.AttributeProto.AttributeType.TENSORS: lambda a, cc: ArrayAttr.get(
+        [cc.tensor_proto_to_attr(t) for t in a.tensors]
+    ),
+    onnx.AttributeProto.AttributeType.GRAPHS: False,
+    onnx.AttributeProto.AttributeType.SPARSE_TENSORS: False,
+    onnx.AttributeProto.AttributeType.TYPE_PROTOS: False,
+}
+
+
+def _get_attr(node: onnx.NodeProto, attr_name: str) -> onnx.AttributeProto:
+    for attr in node.attribute:
+        if attr.name == attr_name:
+            return attr
+    else:
+        raise OnnxImportError(f"Required attribute {attr_name} not found in {node}")

setup.py

@@ -47,8 +47,6 @@ PACKAGE_VERSION = os.environ.get("TORCH_MLIR_PYTHON_PACKAGE_VERSION") or "0.0.1"
 # If true, enable LTC build by default
 TORCH_MLIR_ENABLE_LTC_DEFAULT = True
 TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS = int(os.environ.get('TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS', False))
-if not TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS:
-    import torch
 
 # Build phase discovery is unreliable. Just tell it what phases to run.
 class CustomBuild(_build):
@@ -91,7 +89,7 @@ class CMakeBuild(build_py):
                 f"-DCMAKE_C_VISIBILITY_PRESET=hidden",
                 f"-DCMAKE_CXX_VISIBILITY_PRESET=hidden",
                 f"-DTORCH_MLIR_ENABLE_LTC={'ON' if enable_ltc else 'OFF'}",
-                f"-DTORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS={'ON' if TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS else 'OFF'}",
+                f"-DTORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS={'OFF' if TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS else 'ON'}",
             ]
 
             os.makedirs(cmake_build_dir, exist_ok=True)
@@ -145,8 +143,31 @@ with open("README.md", "r", encoding="utf-8") as fh:
     long_description = fh.read()
 
 
+# Requires and extension modules depend on whether building PyTorch
+# extensions.
+INSTALL_REQUIRES = [
+    "numpy",
+    "packaging",
+]
+EXT_MODULES = [
+    CMakeExtension("torch_mlir._mlir_libs._torchMlir"),
+]
+NAME = "torch-mlir-core"
+
+# If building PyTorch extensions, customize.
+if not TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS:
+    import torch
+    NAME = "torch-mlir"
+    INSTALL_REQUIRES.extend([
+        f"torch=={torch.__version__}".split("+", 1)[0],
+    ])
+    EXT_MODULES.extend([
+        CMakeExtension("torch_mlir._mlir_libs._jit_ir_importer"),
+    ])
+
+
 setup(
-    name="torch-mlir" if not TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS else "torch-mlir-core",
+    name=NAME,
     version=f"{PACKAGE_VERSION}",
     author="Sean Silva",
     author_email="silvasean@google.com",
@@ -159,10 +180,12 @@ setup(
         "built_ext": NoopBuildExtension,
         "build_py": CMakeBuild,
     },
-    ext_modules=[
+    ext_modules=EXT_MODULES,
-        CMakeExtension("torch_mlir._mlir_libs._jit_ir_importer"),
+    install_requires=INSTALL_REQUIRES,
-    ] if not TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS else [CMakeExtension("torch_mlir._mlir_libs._torchMlir")],
+    extras_require={
-    install_requires=["numpy", "packaging"] + (
+        "onnx": [
-        [f"torch=={torch.__version__}".split("+", 1)[0], ] if not TORCH_MLIR_ENABLE_ONLY_MLIR_PYTHON_BINDINGS else []),
+            "onnx>=1.15.0",
+        ],
+    }
     zip_safe=False,
 )

test-requirements.txt

@@ -1,3 +1,4 @@
 pillow
 dill
 multiprocess
+onnx==1.15.0

test/python/lit.local.cfg

@@ -0,0 +1,2 @@
+if not config.enable_bindings_python:
+  config.unsupported = True

test/python/onnx_importer/.gitignore

@@ -0,0 +1 @@
+output/

test/python/onnx_importer/_torch_mlir_config.py

@@ -0,0 +1,19 @@
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Also available under a BSD-style license. See LICENSE.
+
+# RUN: %PYTHON %s
+
+"""This file exists so that the tests can find/configure torch_mlir.
+
+It allows the test file to be standalone and used verbatim in other
+projects (i.e. by just providing this file on the side).
+"""
+
+from torch_mlir import ir
+from torch_mlir.extras import onnx_importer
+
+def configure_context(context):
+    from torch_mlir.dialects import torch as torch_d
+    torch_d.register_dialect(context)

test/python/onnx_importer/import_smoke_test.py

@@ -0,0 +1,374 @@
+# Based on code Copyright (c) Advanced Micro Devices, Inc.
+#
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Also available under a BSD-style license. See LICENSE.
+
+# RUN: %PYTHON %s --output %t
+
+from glob import glob
+from pathlib import Path
+
+import logging
+import sys
+import unittest
+
+import onnx
+
+from _torch_mlir_config import (
+    configure_context,
+    ir,
+    onnx_importer,
+)
+
+# Accept the output path on the command line or default to a sibling
+# to this file. We have to pop this off explicitly or else unittest
+# won't understand.
+if len(sys.argv) > 1 and sys.argv[1] == "--output":
+    OUTPUT_PATH = Path(sys.argv[2])
+    del sys.argv[1:3]
+else:
+    OUTPUT_PATH = Path(__file__).resolve().parent / "output"
+
+
+# TODO: Add some verification and overrides. For now, just use the
+# onnx package install for onnx test files, since they were nice
+# enough to include the test suite in the deployable.
+import onnx.backend.test.data
+
+ONNX_TEST_DATA_DIR = Path(onnx.backend.test.__file__).resolve().parent / "data"
+print(f"ONNX Test Data Dir: {ONNX_TEST_DATA_DIR}")
+ONNX_REL_PATHS = glob(f"**/*.onnx", root_dir=ONNX_TEST_DATA_DIR, recursive=True)
+
+OUTPUT_PATH.mkdir(parents=True, exist_ok=True)
+
+TEST_CAST_XFAILS = [
+    "light_light_bvlc_alexnet",
+    "light_light_inception_v1",
+    "light_light_squeezenet",
+    "light_light_vgg19",
+    "node_test_affine_grid_2d_align_corners_expanded_model",
+    "node_test_affine_grid_2d_expanded_model",
+    "node_test_affine_grid_3d_align_corners_expanded_model",
+    "node_test_affine_grid_3d_expanded_model",
+    "node_test_ai_onnx_ml_label_encoder_string_int_model",
+    "node_test_ai_onnx_ml_label_encoder_string_int_no_default_model",
+    "node_test_ai_onnx_ml_label_encoder_tensor_mapping_model",
+    "node_test_ai_onnx_ml_label_encoder_tensor_value_only_mapping_model",
+    "node_test_cast_FLOAT16_to_FLOAT8E4M3FNUZ_model",
+    "node_test_cast_FLOAT16_to_FLOAT8E4M3FN_model",
+    "node_test_cast_FLOAT16_to_FLOAT8E5M2FNUZ_model",
+    "node_test_cast_FLOAT16_to_FLOAT8E5M2_model",
+    "node_test_cast_FLOAT8E4M3FNUZ_to_FLOAT16_model",
+    "node_test_cast_FLOAT8E4M3FNUZ_to_FLOAT_model",
+    "node_test_cast_FLOAT8E4M3FN_to_FLOAT16_model",
+    "node_test_cast_FLOAT8E4M3FN_to_FLOAT_model",
+    "node_test_cast_FLOAT8E5M2FNUZ_to_FLOAT16_model",
+    "node_test_cast_FLOAT8E5M2FNUZ_to_FLOAT_model",
+    "node_test_cast_FLOAT8E5M2_to_FLOAT16_model",
+    "node_test_cast_FLOAT8E5M2_to_FLOAT_model",
+    "node_test_cast_FLOAT_to_FLOAT8E4M3FNUZ_model",
+    "node_test_cast_FLOAT_to_FLOAT8E4M3FN_model",
+    "node_test_cast_FLOAT_to_FLOAT8E5M2FNUZ_model",
+    "node_test_cast_FLOAT_to_FLOAT8E5M2_model",
+    "node_test_cast_FLOAT_to_STRING_model",
+    "node_test_cast_STRING_to_FLOAT_model",
+    "node_test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FNUZ_model",
+    "node_test_cast_no_saturate_FLOAT16_to_FLOAT8E4M3FN_model",
+    "node_test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2FNUZ_model",
+    "node_test_cast_no_saturate_FLOAT16_to_FLOAT8E5M2_model",
+    "node_test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FNUZ_model",
+    "node_test_cast_no_saturate_FLOAT_to_FLOAT8E4M3FN_model",
+    "node_test_cast_no_saturate_FLOAT_to_FLOAT8E5M2FNUZ_model",
+    "node_test_cast_no_saturate_FLOAT_to_FLOAT8E5M2_model",
+    "node_test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_expanded_model",
+    "node_test_castlike_FLOAT8E4M3FNUZ_to_FLOAT_model",
+    "node_test_castlike_FLOAT8E4M3FN_to_FLOAT_expanded_model",
+    "node_test_castlike_FLOAT8E4M3FN_to_FLOAT_model",
+    "node_test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_expanded_model",
+    "node_test_castlike_FLOAT8E5M2FNUZ_to_FLOAT_model",
+    "node_test_castlike_FLOAT8E5M2_to_FLOAT_expanded_model",
+    "node_test_castlike_FLOAT8E5M2_to_FLOAT_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_expanded_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E4M3FNUZ_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E4M3FN_expanded_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E4M3FN_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_expanded_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E5M2FNUZ_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E5M2_expanded_model",
+    "node_test_castlike_FLOAT_to_FLOAT8E5M2_model",
+    "node_test_castlike_FLOAT_to_STRING_expanded_model",
+    "node_test_castlike_FLOAT_to_STRING_model",
+    "node_test_castlike_STRING_to_FLOAT_expanded_model",
+    "node_test_castlike_STRING_to_FLOAT_model",
+    "node_test_center_crop_pad_crop_axes_chw_expanded_model",
+    "node_test_center_crop_pad_crop_axes_hwc_expanded_model",
+    "node_test_center_crop_pad_crop_negative_axes_hwc_expanded_model",
+    "node_test_clip_default_inbounds_model",
+    "node_test_clip_default_int8_inbounds_model",
+    "node_test_clip_default_int8_max_model",
+    "node_test_clip_default_max_model",
+    "node_test_constantofshape_float_ones_model",
+    "node_test_constantofshape_int_shape_zero_model",
+    "node_test_constantofshape_int_zeros_model",
+    "node_test_dequantizelinear_e4m3fn_model",
+    "node_test_dequantizelinear_e4m3fn_zero_point_model",
+    "node_test_dequantizelinear_e5m2_model",
+    "node_test_dft_axis_model",
+    "node_test_dft_inverse_model",
+    "node_test_dft_model",
+    "node_test_equal_string_broadcast_model",
+    "node_test_equal_string_model",
+    "node_test_gru_defaults_model",
+    "node_test_gru_seq_length_model",
+    "node_test_gru_with_initial_bias_model",
+    "node_test_identity_opt_model",
+    "node_test_identity_sequence_model",
+    "node_test_if_model",
+    "node_test_if_opt_model",
+    "node_test_if_seq_model",
+    "node_test_layer_normalization_2d_axis0_expanded_model",
+    "node_test_layer_normalization_2d_axis0_expanded_ver18_model",
+    "node_test_layer_normalization_2d_axis1_expanded_model",
+    "node_test_layer_normalization_2d_axis1_expanded_ver18_model",
+    "node_test_layer_normalization_2d_axis_negative_1_expanded_model",
+    "node_test_layer_normalization_2d_axis_negative_1_expanded_ver18_model",
+    "node_test_layer_normalization_2d_axis_negative_2_expanded_model",
+    "node_test_layer_normalization_2d_axis_negative_2_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis0_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis0_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis1_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis1_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis2_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis2_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis_negative_1_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis_negative_1_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis_negative_2_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis_negative_2_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_3d_axis_negative_3_epsilon_expanded_model",
+    "node_test_layer_normalization_3d_axis_negative_3_epsilon_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis0_expanded_model",
+    "node_test_layer_normalization_4d_axis0_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis1_expanded_model",
+    "node_test_layer_normalization_4d_axis1_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis2_expanded_model",
+    "node_test_layer_normalization_4d_axis2_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis3_expanded_model",
+    "node_test_layer_normalization_4d_axis3_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis_negative_1_expanded_model",
+    "node_test_layer_normalization_4d_axis_negative_1_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis_negative_2_expanded_model",
+    "node_test_layer_normalization_4d_axis_negative_2_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis_negative_3_expanded_model",
+    "node_test_layer_normalization_4d_axis_negative_3_expanded_ver18_model",
+    "node_test_layer_normalization_4d_axis_negative_4_expanded_model",
+    "node_test_layer_normalization_4d_axis_negative_4_expanded_ver18_model",
+    "node_test_layer_normalization_default_axis_expanded_model",
+    "node_test_layer_normalization_default_axis_expanded_ver18_model",
+    "node_test_loop11_model",
+    "node_test_loop13_seq_model",
+    "node_test_loop16_seq_none_model",
+    "node_test_lstm_defaults_model",
+    "node_test_lstm_with_initial_bias_model",
+    "node_test_lstm_with_peepholes_model",
+    "node_test_optional_get_element_optional_sequence_model",
+    "node_test_optional_get_element_optional_tensor_model",
+    "node_test_optional_get_element_sequence_model",
+    "node_test_optional_has_element_empty_no_input_name_optional_input_model",
+    "node_test_optional_has_element_empty_no_input_name_tensor_input_model",
+    "node_test_optional_has_element_empty_optional_input_model",
+    "node_test_optional_has_element_optional_input_model",
+    "node_test_optional_has_element_tensor_input_model",
+    "node_test_quantizelinear_e4m3fn_model",
+    "node_test_quantizelinear_e5m2_model",
+    "node_test_range_float_type_positive_delta_expanded_model",
+    "node_test_range_int32_type_negative_delta_expanded_model",
+    "node_test_regex_full_match_basic_model",
+    "node_test_regex_full_match_email_domain_model",
+    "node_test_regex_full_match_empty_model",
+    "node_test_resize_downsample_scales_cubic_A_n0p5_exclude_outside_model",
+    "node_test_resize_downsample_scales_cubic_align_corners_model",
+    "node_test_resize_downsample_scales_cubic_antialias_model",
+    "node_test_resize_downsample_scales_cubic_model",
+    "node_test_resize_downsample_scales_linear_align_corners_model",
+    "node_test_resize_downsample_scales_linear_antialias_model",
+    "node_test_resize_downsample_scales_linear_half_pixel_symmetric_model",
+    "node_test_resize_downsample_scales_linear_model",
+    "node_test_resize_downsample_scales_nearest_model",
+    "node_test_resize_downsample_sizes_cubic_antialias_model",
+    "node_test_resize_downsample_sizes_cubic_model",
+    "node_test_resize_downsample_sizes_linear_antialias_model",
+    "node_test_resize_downsample_sizes_linear_pytorch_half_pixel_model",
+    "node_test_resize_downsample_sizes_nearest_model",
+    "node_test_resize_downsample_sizes_nearest_not_larger_model",
+    "node_test_resize_downsample_sizes_nearest_not_smaller_model",
+    "node_test_resize_tf_crop_and_resize_axes_2_3_model",
+    "node_test_resize_tf_crop_and_resize_axes_3_2_model",
+    "node_test_resize_tf_crop_and_resize_model",
+    "node_test_resize_upsample_scales_cubic_A_n0p5_exclude_outside_model",
+    "node_test_resize_upsample_scales_cubic_align_corners_model",
+    "node_test_resize_upsample_scales_cubic_asymmetric_model",
+    "node_test_resize_upsample_scales_cubic_model",
+    "node_test_resize_upsample_scales_linear_align_corners_model",
+    "node_test_resize_upsample_scales_linear_half_pixel_symmetric_model",
+    "node_test_resize_upsample_scales_linear_model",
+    "node_test_resize_upsample_scales_nearest_axes_2_3_model",
+    "node_test_resize_upsample_scales_nearest_axes_3_2_model",
+    "node_test_resize_upsample_scales_nearest_model",
+    "node_test_resize_upsample_sizes_cubic_model",
+    "node_test_resize_upsample_sizes_nearest_axes_2_3_model",
+    "node_test_resize_upsample_sizes_nearest_axes_3_2_model",
+    "node_test_resize_upsample_sizes_nearest_ceil_half_pixel_model",
+    "node_test_resize_upsample_sizes_nearest_floor_align_corners_model",
+    "node_test_resize_upsample_sizes_nearest_model",
+    "node_test_resize_upsample_sizes_nearest_not_larger_model",
+    "node_test_resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric_model",
+    "node_test_rnn_seq_length_model",
+    "node_test_scan9_sum_model",
+    "node_test_scan_sum_model",
+    "node_test_sequence_insert_at_back_model",
+    "node_test_sequence_insert_at_front_model",
+    "node_test_sequence_map_add_1_sequence_1_tensor_expanded_model",
+    "node_test_sequence_map_add_1_sequence_1_tensor_model",
+    "node_test_sequence_map_add_2_sequences_expanded_model",
+    "node_test_sequence_map_add_2_sequences_model",
+    "node_test_sequence_map_extract_shapes_expanded_model",
+    "node_test_sequence_map_extract_shapes_model",
+    "node_test_sequence_map_identity_1_sequence_1_tensor_expanded_model",
+    "node_test_sequence_map_identity_1_sequence_1_tensor_model",
+    "node_test_sequence_map_identity_1_sequence_expanded_model",
+    "node_test_sequence_map_identity_1_sequence_model",
+    "node_test_sequence_map_identity_2_sequences_expanded_model",
+    "node_test_sequence_map_identity_2_sequences_model",
+    "node_test_simple_rnn_defaults_model",
+    "node_test_simple_rnn_with_initial_bias_model",
+    "node_test_split_to_sequence_1_model",
+    "node_test_split_to_sequence_2_model",
+    "node_test_split_to_sequence_nokeepdims_model",
+    "node_test_stft_model",
+    "node_test_string_concat_broadcasting_model",
+    "node_test_string_concat_empty_string_model",
+    "node_test_string_concat_model",
+    "node_test_string_concat_utf8_model",
+    "node_test_string_concat_zero_dimensional_model",
+    "node_test_string_split_basic_model",
+    "node_test_string_split_consecutive_delimiters_model",
+    "node_test_string_split_empty_string_delimiter_model",
+    "node_test_string_split_empty_tensor_model",
+    "node_test_string_split_maxsplit_model",
+    "node_test_string_split_no_delimiter_model",
+    "node_test_strnormalizer_export_monday_casesensintive_lower_model",
+    "node_test_strnormalizer_export_monday_casesensintive_nochangecase_model",
+    "node_test_strnormalizer_export_monday_casesensintive_upper_model",
+    "node_test_strnormalizer_export_monday_empty_output_model",
+    "node_test_strnormalizer_export_monday_insensintive_upper_twodim_model",
+    "node_test_strnormalizer_nostopwords_nochangecase_model",
+    "simple_test_sequence_model1_model",
+    "simple_test_sequence_model2_model",
+    "simple_test_sequence_model3_model",
+    "simple_test_sequence_model4_model",
+    "simple_test_sequence_model5_model",
+    "simple_test_sequence_model6_model",
+    "simple_test_sequence_model7_model",
+    "simple_test_sequence_model8_model",
+    "simple_test_strnorm_model_monday_casesensintive_lower_model",
+    "simple_test_strnorm_model_monday_casesensintive_nochangecase_model",
+    "simple_test_strnorm_model_monday_casesensintive_upper_model",
+    "simple_test_strnorm_model_monday_empty_output_model",
+    "simple_test_strnorm_model_monday_insensintive_upper_twodim_model",
+    "simple_test_strnorm_model_nostopwords_nochangecase_model",
+]
+
+
+class ImportSmokeTest(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.unexpected_failure_count = 0
+        ImportSmokeTest.actual_failures = []
+
+    @classmethod
+    def tearDownClass(cls):
+        if cls.unexpected_failure_count:
+            # Print a helpful message with copy-paste XFAIL def.
+            failure_report_path = OUTPUT_PATH / "import_smoke_test_report.txt"
+            print(
+                "Unexpected failures. Writing copy/paste report to:",
+                failure_report_path,
+            )
+            with open(failure_report_path, "wt") as f:
+                lines = [f'    "{s}",' for s in ImportSmokeTest.actual_failures]
+                print(
+                    f"Unexpected failures in the following. Copy/paste to update `TEST_CAST_XFAILS`:",
+                    file=f,
+                )
+                print(f"TEST_CAST_XFAILS = [", file=f)
+                [print(l, file=f) for l in lines]
+                print(f"]", file=f)
+
+        ImportSmokeTest.actual_failures.clear()
+
+    def load_onnx_model(self, file_path: Path) -> onnx.ModelProto:
+        raw_model = onnx.load(file_path)
+        try:
+            inferred_model = onnx.shape_inference.infer_shapes(raw_model)
+        except onnx.onnx_cpp2py_export.shape_inference.InferenceError as e:
+            print("WARNING: Shape inference failure (skipping test):", e)
+            self.skipTest(reason="shape inference failure")
+
+        # inferred_model = raw_model
+        return inferred_model
+
+    def run_import_test(self, norm_name: str, rel_path: str):
+        context = ir.Context()
+        configure_context(context)
+
+        model_info = onnx_importer.ModelInfo(
+            self.load_onnx_model(ONNX_TEST_DATA_DIR / rel_path),
+        )
+        m = model_info.create_module(context=context)
+        try:
+            imp = onnx_importer.NodeImporter.define_function(model_info.main_graph, m)
+            imp.import_all()
+            m.verify()
+        finally:
+            # Use a ".txt" extension to avoid lit test discovery.
+            with open(OUTPUT_PATH / f"{norm_name}.mlir", "wt") as f:
+                print(m.get_asm(), file=f)
+
+    def testExists(self):
+        # We expect a lot of test cases. Die if not the case (i.e. if paths change
+        # or something).
+        self.assertGreater(len(ONNX_REL_PATHS), 10)
+
+
+# Generate test methods for each onnx file.
+for _rel_path in ONNX_REL_PATHS:
+
+    def attach_test(rel_path):
+        norm_name = rel_path.removesuffix(".onnx").replace("/", "_")
+
+        def test_method(self: ImportSmokeTest):
+            try:
+                self.run_import_test(norm_name, rel_path)
+            except onnx_importer.OnnxImportError as e:
+                # All legitimate failures should be caught and reported
+                # as an OnnxImportError.
+                ImportSmokeTest.actual_failures.append(norm_name)
+                if norm_name not in TEST_CAST_XFAILS:
+                    ImportSmokeTest.unexpected_failure_count += 1
+                raise e
+
+        test_method.__name__ = f"test_{norm_name}"
+
+        if norm_name in TEST_CAST_XFAILS:
+            test_method = unittest.expectedFailure(test_method)
+
+        setattr(ImportSmokeTest, test_method.__name__, test_method)
+
+    attach_test(_rel_path)
+
+
+if __name__ == "__main__":
+    logging.basicConfig(level=logging.DEBUG)
+    unittest.main()

test/python/onnx_importer/lit.local.cfg

@@ -0,0 +1,5 @@
+try:
+  import onnx
+except ModuleNotFoundError:
+  print("Skipping onnx tests.. no onnx")
+  config.unsupported = True